Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Optimal Option Portfolio Strategies: Deepening the Puzzle of Index Option Mispricing José Afonso Faias and Pedro Santa-Clara∗ ∗ Faias, [email protected], UCP - Católica Lisbon School of Business & Economics, Palma de Cima, 1649-023 Lisboa, Portugal; Santa-Clara, [email protected], Millennium Chair in Finance. Nova School of Business and Economics, NBER, and CEPR, Campus de Campolide, 1099-032 Lisboa, Portugal. We thank the editor Hendrik Bessembinder and an anonymous referee for valuable comments which significantly improved the paper. We also thank Rui Albuquerque, Pierre CollinDufresne, Joost Driessen, Miguel Ferreira, Mark Grinblatt, José Correia Guedes, Christopher Jones, Ángel León, André Lucas, Pedro Matos, David Moreno, Andreas Rathgeber, Enrique Sentana, Ivan Shaliastovich, and participants at Nova School of Business and Economics, the QED 2010 Meeting at Alicante, Católica Lisbon School of Business and Economics, the 6th PFN Conference, the Finance & Economics 2010 Conference, the XVIII Foro de Finanzas, the 2011 AFA Annual Meeting, the 5th Conference on Professional Asset Management, the Oxford–Man Institute at Quantitative Finance, the 2011 EFMA Annual Meeting, the EcoMod2011, the Conference in honor of Richard Roll and Eduardo Schwartz, the Swissquote Conference 2011, the Sixth Annual Conference on Advances in the Analysis of Hedge Fund Strategies, the Boston University, and the Research in Options Conference for helpful comments and discussions. This research was funded by grant PTDC/EGE-ECO/119683/2010 of the Portuguese Foundation for Science and Technology-FCT. Abstract Traditional methods of asset allocation (such as mean-variance optimization) are not adequate for option portfolios because the distribution of returns is non-normal and the short sample of option returns available makes it difficult to estimate their distribution. We propose a method to optimize a portfolio of European options, held to maturity, with a myopic objective function that overcomes these limitations. In an out-of-sample exercise, incorporating realistic transaction costs, the portfolio strategy delivers a Sharpe ratio of 0.82 with positive skewness. This performance is mostly obtained by exploiting mispricing between options and not by loading on jump or volatility risk premia. I Introduction Although options are well known to help span states of nature (Ross (1976)) and to provide exposure to (priced) risk factors like stochastic volatility and jumps,1 they are seldom used in investment portfolios.2 Part of the problem is that the portfolio optimization methods available, like the Markowitz mean-variance model, are ill suited to handle options. There are three main problems in option portfolio optimization. First, the distribution of option returns departs significantly from normality and therefore cannot be described by means and (co)variances alone. Second, the short history of option returns available severely limits the precision of the estimation of their complex joint distribution thereby adding considerable estimation error to the portfolio.3 For example, we have only data for Standard & Poor’s 500 options since 1996, which is not long enough to estimate reliably the moments of their joint return distribution. Third, there are high transaction costs in this market. On average, at-the-money (ATM) options have a 5% relative bid-ask spread, while out-of-the-money (OTM) options have relative bid-ask spreads of 10%. Taking these costs into account is thus critical but unfortunately hard to do within the traditional optimization approach. We offer a simple portfolio optimization method – optimal option portfolio strategies (OOPS) – that solves these problems. Instead of a mean-variance objective, we maximize 1 See Bates (1996), Bakshi, Cao, and Chen (1997), Andersen, Benzoni, and Lund (2002), and Liu and Pan (2003), among others. 2 Mutual funds’ use of derivatives is limited (Koski and Pontiff (1999), Deli and Varma (2002), Almazan, Brown, Carlson and Chapman (2004)). Mutual funds generally face legal constraints in terms of shortselling, borrowing and derivatives usage. This does not happen with hedge funds, which do use derivatives, but these represent only a small part of their holdings (Chen (2011), Aragon and Martin (2012)). 3 This is especially problematic since the optimization process tends to amplify this estimation noise (Michaud and Michaud (2008)). 1 an expected utility function, such as power utility, which takes into account all the moments of the portfolio return distribution, penalizing negative skewness and high kurtosis. We deal with the limited sample of option returns by relying only on historical data for the underlying asset. The only option data needed to simulate option returns are their current prices, as the payoff is fully determined by the simulations of the underlying asset. We simulate returns of the underlying asset going forward and, given the definition of option payoffs, simulate corresponding option returns. We use simple bootstrap methods fitted to index data since 1950. Plugging the simulated returns of the option portfolio into the utility function and averaging across simulations gives us an approximation of the expected utility that can be maximized to obtain optimal portfolio weights.4 This metholodogy is similar in spirit to Brandt, Santa-Clara, and Valkanov (2009). We apply our method to a portfolio allocation problem between a risk-free asset and four European options on the S&P 500 index with one-month to maturity: a ATM call, a ATM put, a 5% OTM call, and a 5% OTM put option.5 These are liquid options that can be combined to generate a variety of final payoffs. We study optimal option strategies in a realistic setting with transaction costs. To incorporate transaction costs, we follow Eraker (2013) and Plyakha and Vilkov (2008). For each option, we define two securities: a “long option” initiated at the ask quote and a “short option” initiated at the bid quote. The “short options” enter the optimization process with negative signal. Finally, we impose no 4 Hedging demands and long-term horizons might potentially be important. However, Tan (2013) finds that the benefit of adding options is quite small for long-horizon constant relative risk aversion (CRRA) investors and Liu and Pan (2003) and Driessen and Maenhout (2007) show that improvements by including derivatives are driven mostly by a myopic component. Our results show low predictability of returns of the optimal strategy and little correlation with the stock market. These two features imply that there would be no hedging demand and thus little loss from our myopic approach. 5 The method is usable with American options if we are willing to ignore the impact of early exercise. It is an empirical question whether the method remains useful. 2 short-selling so that only one of the two, long or short positions, is ever taken. We study the performance of the OOPS in an out-of-sample exercise using a conservative CRRA utility function. Between January 1996 and August 2013, the OOPS yields an annualized certainty equivalent of 9.94% and an annualized Sharpe ratio of 0.82. This compares well with the stock market in the same period which had a certainty equivalent of −1.31% and a Sharpe ratio of 0.29. Surprisingly, OOPS returns present positive skewness and relatively low excess kurtosis which compare favorably to simply shorting individual options. We find that our strategy loads significantly on all four options and that the optimal weights vary over time. OOPS is almost delta-neutral and has a low beta of 0.03 on average. This is interesting as we expected a greater exposure to the stock market given that the return distribution on which the portfolio choice is based is calibrated to data since 1950 in which the equity Sharpe ratio was quite large at around 0.48. Besides loading on the equity premium, there could be two reasons for an investor to buy or sell options: to obtain exposure to risk premia for volatility and jump risk or to exploit option mispricing. Broadie, Chernov, and Johannes (2009) argue that there is little evidence of mispricing of options relative to the underlying stock index. However, we show that the OOPS returns are little exposed to the market, volatility and jump risk factors identified in the literature as having significant premia implying that the main source of returns come from exploiting mispricings of options relative to each other. A related literature investigates the returns of simple option trading strategies. Coval and Shumway (2001) show that short positions in crash-protected, delta-neutral straddles present Sharpe ratios of around one.6 Driessen and Maenhout (2013) confirm these results for short-term options in U.S. and U.K. markets. Coval and Shumway (2001) and Bondarenko (2003) also find that selling naked puts offers high returns even after taking into account their considerable risk. In contrast to OOPS, all these strategies have negative skewness 6 The sample period of their study is different from ours making the comparison of Sharpe ratios impos- sible. 3 and high kurtosis implying a lower (negative in many cases) certainty equivalent than ours. We find that our portfolio departs significantly from exploiting these simple strategies. For instance, there are several periods in which the OOPS is net long in options. There are a few papers that also address optimal portfolio allocation with options. Liu and Pan (2003) model stochastic volatility and jump processes and derive the optimal portfolio policy of a CRRA investor across one stock, a 5% OTM put option, and cash. Although they obtain an analytic solution for the optimal option allocation, they need to specify a particular parametric process for the stock dynamics and estimate its parameters. They try different parameter sets and obtain ambiguous conclusions in terms of optimal put weights. Most importantly, their approach cannot be extended to multiple options. In contrast, we can use any model (parametric or not) for the distribution of returns of the underlying asset. Our work is empirical in nature and we impose no restrictions on the number of options that could be used in the optimization problem. Driessen and Maenhout (2007) maximize the average utility of realized returns for a portfolio of a stock, a simple option strategy, and cash. As the option strategy, they consider either a 4% OTM put, a ATM straddle, or corresponding crash-neutral strategies. They conclude that positive put holdings are never optimal given historic option returns. We find, in contrast, that optimal option weights are time-varying and change signs during our sample period. Jones (2006) studies optimal portfolios to exploit the apparent put mispricing. He uses a general nonlinear latent factor model and maximizes a constrained mean-variance objective. He circumvents the short history of data by using daily option returns. However, his model is quite complex, with 57 parameters to estimate even when only one factor is considered. This limits the practical usefulness of this approach. Constantinides, Jackwerth, and Savov (2013) study portfolios made up of either calls or puts with a targeted moneyness. Although they find high Sharpe ratios, mostly for put strategies, these strategies yield negative skewness and high kurtosis implying a lower (or 4 even negative) certainty equivalent than the OOPS. Malamud (2014) studies dynamic portfolio selection with out-of-the-money options. His approach, based on option greeks, allows him to combine several options to obtain high Shape ratios albeit with substantial kurtosis. His approach induces large turnover which leads to very high transaction costs. The paper is organized as follows. Section II explains the methodology. Section III describes the data used. Section IV presents the results. Finally, we present some concluding remarks. II Portfolio Allocation A Methodology Let time be represented by the subscript t and simulations indexed by n. Our portfolio allocation is implemented for one risk-free asset and a series of call and put options with one period to maturity. We consider call options indexed by c and put options indexed by p. We include only assets that are not redundant by put-call parity.7 At time t, the value of the underlying asset is denoted by St and each option i has an exercise price of Kt,i . The risk-free interest rate from time t to t + 1, known at time t, is denoted by rft . For each date t, weights are obtained through maximization of the investor’s expected utility of end-of-period wealth, which is a function of simulated portfolio returns. The latter are derived from option returns, which in turn depend on the underlying asset returns. The steps below describe the OOPS algorithm in detail. Appendix A shows a simple illustration. n 1. We simulate N log-returns rt+1 , n = 1, ..., N of the underlying asset. The simulation is performed under the empirical density not the risk-adjusted measure. 2. The returns from step 1 are used to simulate the next period’s underlying asset value, 7 Although put-call parity does not hold because of transactions costs, incorporating the stock market is still redundant because its payoff is strickly dominated by the other assets’ payoffs. 5 given its current value: n n St+1|t = St exp(rt+1 ) (1) n where n = 1, ..., N , and St+1|t denotes the simulated underlying asset value in period t + 1 conditional on information up to time t, and St denotes the current underlying asset value, which is observed. 3. Using the known strike prices for call options, Kt,c , and put options, Kt,p , and oneperiod simulated underlying asset values St+1|t from Equation (1), we simulate option payoffs at their maturity t + 1: (2) n n n n Ct+1|t,c = max(St+1|t − Kt,c , 0) and Pt+1|t,p = max(Kt,p − St+1|t , 0) where n = 1, ..., N . Using the simulated payoffs in Equation (2) and observed current option prices Ct,c and Pt,p , option returns are computed by: n rt+1|t,c (3) = n Ct+1|t,c Ct,c − 1 and n rt+1|t,p = n Pt+1|t,p Pt,p −1 where n = 1, ..., N . 4. We construct simulated portfolio returns in the usual way: (4) rpnt+1|t = rft + C X n ωt,c (rt+1|t,c c=1 − rft ) + P X n ωt,p (rt+1|t,p − rft ) p=1 where ωt,c and ωt,p denote the call and put option weights and n = 1, ..., N . Each simulated portfolio return is a weighted average of the asset returns and only the risk-free rate is not simulated. 5. We choose weights by maximizing expected utility over simulated portfolio returns (5) N 1 X max E U Wt [1 + rpt+1|t ] ≈ max U Wt [1 + rpnt+1|t ] . ω ω N n=1 6 The output is given by ωt,c and ωt,p . 6. One-period out-of-sample performance is evaluated with realized option returns. We determine the option realized payoffs: (6) Ct+1,c = max(St+1 − Kt,c , 0) and Pt+1,p = max(Kt,p − St+1 , 0) We find the corresponding returns: (7) rt+1,c = Pt+1,p Ct+1,c − 1 and rt+1,p = −1 Ct,c Pt,p Finally, we determine the one-period out-of-sample portfolio return: (8) rpt+1 = rft + C X ωt,c (rt+1,c − rft ) + c=1 P X ωt,p (rt+1,p − rft ) p=1 using the weights determined ex ante in step 5 above. Our approach essentially compares the current price of an option to the expected payoff and risk of the option that follows from the assumed probability distribution of the underlying asset implicit in the simulation. If the option appears cheap relative to the expected payoff and risk profile, the investor buys it. The investor sells the option if its current price is too high relative to the expectation and risk profile of the simulated payoff at maturity. B Return simulation Our method requires the simulation of the distribution of the underlying stock market returns. It is known that the distribution of raw stock market returns is left skewed, leptokurtic, and presents ARCH effects which can be verified in Table 1. We study historic standardized returns, sr, as the ratio between raw returns, rr, and their standard deviation, sdev, as: (9) n srt+1 n rrt+1 = , n = 1, ..., N sdevt+1 From Table 1, we confirm that this series is closer to a normal distribution, although still left-skewed, and no significant ARCH effects are present. 7 Each month after 1995, based on an expanding sample of historical standardized returns starting in 1950, we simulate different paths for standardized returns. Following Efron and Tibshirani (1993), we resample directly from the standardized empirical distribution (bootstrap). This is appropriate since standardized returns are approximately i.i.d. This approach has the same moments as the historical distribution by construction. The final step is to scale up or down simulated returns by the current realized volatility: n n × rvt , n = 1, ..., N = srt+1 rt+1 (10) The current realized volatility is calculated from the last d trading days and scaled by 21 days ( the average number of trading days in a month) to get monthly units. We consider alternatives values of d to be 1, 5, 10, 20, 30, and 60 days. In each month, we only use the realized volatility length that maximizes the expected utility in sample.8 This simple approach to incorporate time-varying volatility is close in spirit to the filtered historical simulation of Barone-Adesi, Giannopolos, and Vosper (1999) in which volatility is estimated by a parametric method such as a generalized autoregressive conditional heteroscedastic (GARCH) model instead of using past realized volatility as we do. Hull and White (1998), Diebold, Schuermann, and Stroughair (1998) and Barone-Adesi, Engle, and Mancini (2008) also build on the same idea. Of course this model is simplistic and we ignore all the great work on GARCH, implied volatility, MIDAS, and time-varying crashes. Our point is to show that our method works well even in a simple setting such as this one. C Maximizing expected utility In the fourth step, the investor maximizes the conditional expected utility of next period’s wealth: 8 We also consider other finer grids and a range-based integrated volatility estimator. The results lead to the same economic conclusions. 8 (11) max ωt,i ∈R E[U (Wt+1 )] subject to the usual budget constraint Wt+1 = Wt (1 + rpt+1 ). Maximizing expected utility takes into account all the moments of the return distribution. If returns are normal, investors care only about the mean and variance of portfolio returns. In practice, normality does not hold, especially for option returns. Investors thus care also about tail risk (extreme events) and asymmetry, so mean and variance do not provide enough information to make asset allocation choices. We use the power utility function (see Brandt (1999)). This utility function presents constant relative risk aversion (CRRA) and is given by: (12) U (W ) = 1 W 1−γ , 1−γ ln(W ), if γ 6= 1 if γ = 1 where γ is the coefficient of relative risk aversion. This utility function is attractive for two reasons. First, because of the homotheticity property, portfolio weights are independent of the initial level of wealth. So maximizing E[U (Wt+1 )] is the same as maximizing E[U (1 + rpt )]. Second, investors care about all moments of the distribution and this particular utility function penalizes negative skewness and high kurtosis.9 Bliss and Panigirtzoglou (2004) estimate an empirical risk aversion of 4 for S&P 500 index option data. Even though, we set the constant relative risk aversion parameter γ equal to 10 in the optimization for a conservative asset allocation choice.10 Intuitively, we optimize 9 We use the routine FMINCON with the active-set algorithm in Matlab. The results are robust to different initial values. 10 We run OOPS for different values of risk aversion coefficient such as 7 and 13 and the performance is similar. 9 as if the risk aversion coefficient was higher than it truly is in reality to minimize in-sample overfitting. We can think of this as akin to shrinking portfolio weights which is often done in practice.11 Finally, note that we can use any other objective function in applying our methodology.12 D Transaction costs There is a large body of literature that documents high transaction costs in the option market that are in part responsible for pricing anomalies such as violations of the put-call parity relation.13 Hence, it is essential to take these frictions into account in our optimization approach. We discuss only the impact of transaction costs measured by the bid-ask spread. Other types of costs like brokerage fees and market price impact may be substantial but are ignored here. Figure 1 shows bid-ask spreads for options between Junuary 1996 and August 2013. Table 2 shows average bid-ask spreads of about $1.40 for ATM options and $0.70 for OTM options. Dividing this by midprices, we measure relative bid-ask spreads for ATM options of around 5%, increasing to 10% on average for OTM options. Relative bid-ask spreads change over time and for OTM options can reach up to 30%.14 We incorporate transaction costs into our approach by decomposing each option into 11 When the optimization leads to extreme portfolio weights – due to the large simulated returns from current option prices – we use the risk-free rate in that month. We do this for realism since in practice options traders face constraints in terms of the risk exposure of their portfolios. We also tried simply using the allocation of the previous month with virtually the same results. 12 Driessen and Maenhout (2007) discuss several alternatives for disappointment aversion. 13 See, for instance, Phillips and Smith (1980), Baesel, Shows, and Thorp (1983), and Saretto and Santa- Clara (2009). 14 Dennis and Mayhew (2009) shows that the effective spread is about two-thirds of the quoted spread given in OptionMetrics, so quoted spreads may overestimate costs which seems a conservative assumption to use in our strategy. 10 two securities: a “bid option” and an “ask option.” We initiate long positions at the ask quote and short positions at the bid quote; shorts enter the optimization with a minus sign. This follows the approach taken by Eraker (2013) and Plyakha and Vilkov (2008). Then we run the algorithm as a constrained optimization problem by imposing a no short-selling condition. This means that in each month only one of the securities, either the bid or the ask option, is ever traded. Note that the wider the bid-ask spread, the less likely an allocation to the security is. III A Data Securities We analyze the optimal option portfolio allocation from January 1996 through August 2013. The choice of the period relates to the availability of option data. We use log returns of the Standard & Poor’s 500 index from February 1950 through August 2013 as the basis for the simulation process. Figure 2 presents the monthly time-series of Standard & Poor’s 500 and VIX indices in the portfolio allocation period.15 This is a rich period with a variety of market conditions comprising events such as the 1997 Asian crisis, the 1998 Russian financial crisis, the 1998 collapse of Long Term Capital Management, the 2000 Nasdaq peak, the September 2001 attack on New York’s World Trade Center, the 2002 business corruption scandals (Enron and Worldcom), the second Gulf War, the 2006-2008 subprime mortgage crisis and Lehman Brothers default, the 2010 sovereign crisis and Flash Crash, and the August 2011 stock market fall. We use data from the OptionMetrics Ivy DB database for European options on the S&P 15 VIX index is calculated and disseminated by the Chicago Board Options Exchange. It estimates the implied volatility of ATM options on the S&P 500 index over the next month. The formula uses a kernelsmoothed estimator that takes as inputs the current market prices for several call and put options over a range of moneyness and maturity. 11 500 index traded on the CBOE.16 The underlying asset is the index level multiplied by 100. Option contracts expire on the third Friday of each month. The options are settled in cash on the business day following expiration. The dataset includes the closing bid and ask prices, volume, and open interest for the period between January 1996 and August 2013. In order to eliminate unreliable data, we apply a series of filters typically used in the literature. First, we exclude all observations for which the bid is lower than $0.125 or higher than the ask price. Second, we exclude all observations with no volume to mitigate the impact of non-trading. Finally, we exclude all observations that violate arbitrage bounds. We assume the risk-free interest rate is represented by the one-month T-Bill. This series is from Bloomberg for the period between January 1996 and August 2013. B Construction of option returns Our asset allocation uses a risk-free asset and a set of risky “securities.” We consider four options with different levels of moneyness: a ATM call, a ATM put, a 5% OTM call, and a 5% OTM put. This limited number of securities keeps the model simple but still generates flexible payoffs as a function of the underlying asset price: OTM options are important for kernel spanning (Buraschi and Jackwerth (2001) and Vanden (2004)) and a deep OTM put option is very sensitive to crash risks.17 The options we use are among the most liquid contracts, using volume as a proxy for liquidity. We choose one-month to maturity options. Buraschi and Jackwerth (2001) report that 16 The options trade under the ticker SPX. Average daily volume in January 2013 was about 800,000 contracts. 17 In practice, you can use more options across moneyness or options across different assets. As an example, we consider three call options (ATM, 2% OTM, and 7% OTM) and three put options (ATM, 2% OTM, and 7% OTM) and the performance is very similar. The only numerical constraint that sometimes occurs is due to redundancy of some options in some periods. This happens when moneyness of the different options values are not sufficiently apart. The problem of a portfolio with a large number of options is incurring high transaction costs prevalent in this market. This solution is thus, in practical terms, infeasible. 12 most of the trading activity in S&P 500 index options is concentrated in the nearest contracts of less than 30 days to expiration. This target maturity is also appealing as longer-maturity option contracts may stop trading if the underlying asset moves in such a way that the options become very deeply ITM or OTM. Most importantly, holding the options to maturity incurs transaction costs only at inception of the trade. We first find all available option contracts with exactly one-month to maturity, i.e., we use the returns from Friday before the third Saturday to Friday before the third Saturday of the next month.18 We then define buckets for option moneyness in terms of the ratio of the underlying price to the strike price less one, S/K − 1. We set a range of moneyness between −1.0% and 1.0% for ATM options and a range of 2% around 5.0% for OTM options. For each month and bucket we are left with several potential securities from which we choose the option with the lowest relative bid-ask spread. When more than one contract has the same spread, we choose the one with the highest open interest. Finally, we construct the synthetic one-month hold-to-expiration option returns: (13) rt,t+1 = Payofft+1 −1 Pricet where Payofft+1 is the payoff of the option at maturity calculated using the closing price of the underlying asset on the day before settlement, and Pricet is the option price observed at the beginning of the period. We obtain a time-series of 212 observations for each security. Figure 3 presents histograms of each individual option returns. For any of the options considered, we confirm that the option return distributions depart significantly from the normal distribution with considerable negative tail risk. Mean time-series characteristics for each option by moneyness are presented in Table 2. ATM call and put options have average 18 There are other alternatives that we do not follow. For example, Buraschi and Jackwerth (2001), Coval and Shumway (2001), and Driessen and Maenhout (2013) select options on the first day of each month and compute returns until the first day of the next month. 13 moneyness of 0.24% and −0.10%, respectively, while OTM call and put options have average moneyness of −4.09% and 4.37%, respectively. These numbers show how close each contract is to the target of each bucket. Volume for each contract is around 13,000 and open interest is close to 15 million. Mean implied volatility varies with moneyness from 17% to 22%, confirming the well known smile effect. Table 3 reports summary statistics of returns for the various securities. We compute the certainty equivalent of an investor with a CRRA utility function with risk aversion parameter of 4, which is given by CE = [(1 − γ)Ū ]1/(1−γ) − 1 (14) where γ is 4, Ū = 1 T PT t=1 Ut and Ut is the CRRA utility function for each month t. We use a risk aversion coefficient of 4 to evaluate the returns of the strategy. This is lower than the coefficient used in the optimization. Intuitively, we optimize as if the risk aversion coefficient was higher than it truly is in reality to minimize in-sample overfitting. We can think of this as akin to shrinking portfolio weights which is ofter done in practice. We also compute the Sharpe ratio, and offer descriptive statistics of the return distribution, in particular, skewness and excess kurtosis.19 We use the S&P 500 total return index as a benchmark.20 The S&P 500 total return index has an average annualized return of 6.4% 19 The main problem of the Sharpe ratio in this setup is that it takes into account only the first two moments, the mean and the standard deviation. See Bernardo and Ledoit (2000) and Ingersoll, Spiegel, Goetzmann and Welch (2007) for problems with the Sharpe ratio. Broadie et al. (2009) show that although the Sharpe ratio is not the best measure to evaluate performance in an option framework, other alternative measures such as Leland’s alpha or the manipulation–proof performance metric face the same problems. 20 We also construct a portfolio consisting of the stock market and the T-bill optimized each month using the CRRA bootstrap methodology as defined in Section II. The performance is similar although with lower and less volatile returns. 14 over the sample period, corresponding to an annualized certainty equivalent of −1.31% and an annualized Sharpe ratio of 0.29.21 This Sharpe ratio is lower than for the entire sample, reflecting the low returns in the recent period. S&P 500 index returns are negatively skewed and leptokurtic. The middle panel of the same table presents the same summary statistics for individual options. Options present large negative average annualized returns ranging from −440% to −131%, except the ATM call with a positive mean return of 52%. This suggests that writing options would have been a good strategy, with annualized Sharpe ratios ranging from 0.14 to 0.60 for this period.22 Writing options, however, has negative tail risk that is very onerous in some months. The returns to writing options have a maximum of 100%, but the minimum ranges from −2, 496% to −664% depending on the option. This leads to large negative skewness and excess kurtosis as high as 38 and a deeply negative certainty equivalent of -100%. The last row of this panel shows a strategy that allocates the same weight to each option. DeMiguel, Garlappi, and Uppal (2009) argue that a naive 1/N uniform rule is generally a good strategy. Using this rule for our four risky assets, we obtain smoother skewness and kurtosis, and an annualized Sharpe ratio of 0.47. However, as with individual options, the certainty equivalent is still −100%. We note that looking at these extremely levered securities may lead to unreasonable returns. No investor would invest all wealth on them. We therefore construct low-leverage portfolios that short only 4.47% of the wealth in each option and go long 104.47% in the risk-free asset. This choice allows us to relate the returns of individual options to our OOPS returns later on since OOPS invests, on average, 104.47% in the risk-free asset and has an average short position of 4.47% in options. Of course, mean, standard deviation, minimum and maximum drop dramatically while skewness, excess kurtosis and annualized Sharpe ratio do not. However, the simple 21 This negative certainty equivalent is due to the low return, high volatility, and large drawdowns of the market in this period, as well as the high risk aversion of our investor. 22 Coval and Shumway (2001) and Eraker (2013) show that writing put options earns Sharpe ratios of 0.68 or above for a different time period. 15 option strategies still deliver negative skewness returns and in most cases negative certainty equivalents showing that risk-averse investors would not want to use them. IV A Results Out-of-sample OOPS returns Figure 4 presents the distribution of OOPS out-of-sample returns. In contrast to individual options, the OOPS is closer to a normal distribution with a symmetric shape and low tail risk. Table 4 reports summary statistics of out-of-sample returns for the OOPS and the S&P 500 total return index (for comparison) between January 1996 and August 2013. Our strategies deliver an annualized certainty equivalent of 9.94% after transaction costs which contrasts to the S&P 500 certainty equivalent of −1.31%. The OOPS yields an average annualized return of 16.1% and an annualized standard deviation of 18.4%. This implies an annualized Sharpe ratio of 0.82.23 The S&P 500 total index returns present lower mean and larger standard deviation which imply a relatively lower annualized Sharpe ratio in this period of 0.29. In contrast to the stock market, OOPS has positive skewness and low kurtosis. This is remarkable since shorting individual options has large negative skewness and much higher kurtosis. The OOPS has a minimum return of −18.4% and a maximum return of 20.8% which compares to the S&P 500 index minimum of −29.5% and a maximum of 18.6%. Figure 5 presents the cumulative returns for OOPS, S&P 500 total return index and risk-free rate starting from $100 invested in January 1996. OOPS shows a relatively stable increasing pattern. The final value is more than thirteen times the initial amount invested, well above the performance of the S&P 500 index.The largest OOPS loss (-18%) was associated with the 9/11 terrorist attacks. The second largest OOPS loss (−15%) was associated with the Flash Crash on May 6, 2010. We cannot relate the third largest OOPS 23 We tried our method for 2-month and 3-month options. For 2-month options, our method delivers a Sharpe ratio of 0.81 and a certainty equivalent of 4.91%. For 3-month options, our method delivers a Sharpe ratio of 0.42 and a certainty equivalent of 8.14%. 16 loss to any special event. On the other side, March 2009 was the inflection point after the 2007-08 crash. May 2013 and February 1996 did not have salient events, although associated with largest OOPS gains. An investor may be worried by OOPS return drawdowns. Nevertheless, the OOPS return drawdowns are typically less severe than drawdowns for the S&P 500 index returns. This can also be confirmed by the positive certainty equivalent of OOPS returns when compared to the negative certainty equivalent of the stock market returns. Another concern would be neglecting price impact costs for OTM puts since the market is fairly one-sided for this option. To test this concern, we run our method but we remove the possibility of selling OTM puts while still allowing them to be bought. Alternatively, we also run the methodology when the bid-ask spread for OTM puts is doubled. In both cases, the impact on performance is marginal since the strategy does not meaningfully short these options. These results are provided in the internet appendix. The OOPS is quite different from the simple short put and other strategies described in the literature (Coval and Shumway (2001), Saretto and Santa-Clara (2009)). Table 6 reports the average, the minimum and the maximum net weights of each option in the strategy and Figure 6 presents the evolution of net weights for each option. Optimal call and put option weights vary significantly over time, ranging from −8.7% to 3.5%. Put options have generally lower weights relative to call options. On average, the OOPS takes low short positions in options. Nonetheless, there are extended periods in which the OOPS is long in options. There is an offsetting effect within ATM options and within OTM options.24 The bottom right picture in Figure 6 describes the evolution of the risk-free security weight which is 100% minus the sum of the weights invested in options. The mean risk-free weight is 104.47% ranging from 94% to 116%. In line with our expectation, OOPS is a net seller of options on average, but in 27% of the months our investor holds a net long position (on average equal to 3%). Note that Driessen and Maenhout (2007) show that constant 24 Correlation figures confirm these results. The strongest correlated pair of weights is the ATM call and puts of the order of 0.82. 17 relative risk aversion investors always find it optimal to short OTM puts, and only with distorted probability assessments using cumulative prospect theory and anticipated utility are they able to obtain positive weights for puts. In contrast, we find that there are many periods when this does not apply. For example, in July 1997, the OOPS takes a long position of 0.7% in OTM put options. In their setting, they have a static setting - constant weights in each option over time – and a smaller opportunity set – only one option strategy at a time. It is clearly not the case that the OOPS methodology simply sells volatility at all times. There is a strong correlation, 0.94, between the sum of call weights and the sum of put weights. Figure 6 shows that the optimal portfolio is short OTM puts most of the time (85% of the months), but partially hedges this position with other options. 1 Choice of volatility estimator OOPS uses the current realized volatility which is calculated from the last d trading days and scaled by 21 days (the average number of trading days in a month) to express it in monthly units. In each month, we use the length d that maximizes the expected utility in sample. Alternatively, we also use the VIX, the RiskMetrics estimate or a fixed lenght of d for computing realized volatility. The RiskMetrics model assumes that the variance process 2 2 where 0 ≤ λ ≤ 1 is the smoothing parameter. The + (1 − λ)rt−1 is given by σt2 = λσt−1 smaller the smoothing parameter, the greater weight is given to recent return data. We present the results of using these estimators for OOPS in Table 5. OOPS returns are robust to the choice of volatility estimator, although there is some deterioration of performance when using VIX, RiskMetrics and different fixed levels for d. The decrease in performance from using VIX shows that part of OOPS returns are due to an underestimation of volatility immediately following a positive shock to volatility. Alternatively, it may be that VIX overshoots the true volatility. The results using RiskMetrics are also enlightening. With higher λ and stronger persistence of the estimator, we get worse performance of OOPS. This means that the performance of our method depends on using very recent information to estimate volatility. We see the same results as we vary the length of the window to 18 estimate volatility. It is particularly important for our method to put most weight on recent observations. B Risk or mispricing? There are two alternative justifications for the OOPS performance: loading in priced risk factors or exploiting option mispricing. The existing literature supports both hypotheses. Bakshi et al. (1997), Bates (1996), Chernov and Ghysels (2000), Duffie, Pan, and Singleton (2000), Pan (2002), Liu and Pan (2003), Eraker (2004), Broadie, Chernov, and Johannes (2007), Todorov (2010), and Christoffersen, Jacobs, and Heston (2013) show that option prices reflect priced volatility and crash risk exposures, whereas Coval and Shumway (2001), Jackwerth (2000), Jones (2006), Saretto and Goyal (2009), and Constantinides et al. (2013) show that risk factor exposures do not fully explain option prices. In this section, we explore the two alternatives. We find that the OOPS has low exposure to priced risk factors. Ex ante, our strategies have, on average, low exposure to the market, volatility and jump risks. There are, however, some periods of time in which OOPS is more exposed, especially when volatility is low. Figure 7 presents the time-series evolution of three risk measures in the period between January 1996 and August 2013. We use greeks from Black and Scholes (1973) to evaluate this. The beta of an option is computed by multiplying the Black–Scholes delta by the ratio between the underlying asset value and the option value. The portfolio beta is the weighted average of the option betas.25 This measure has the advantage that it takes into account option leverage. OOPS has lower beta than individual options.26 It varies between −0.75 and 0.62 with a mean value of 0.03. This means beta is not significantly different from zero using a Newey-West (1987) t-statistic. 25 We do not present the delta, since its time-series distribution is very narrow and close to zero. 26 Beta for individual options is presented in Table 2 and the differences between individual option betas and OOPS betas are statistically significant. 19 We use percent vega to proxy for the volatility exposure. Percent vega is the percentage change in the portfolio due to a one percentage point change in volatility. This is generally low, varying between −1.4% and 0.7%. The mean percent vega is −0.3% and significantly different from zero using a Newey-West t-statistic. So, if volatility increases by 1%, the portfolio returns decrease by 0.3% on average. The magnitude of the exposure is low (and the sign changes often) so that the OOPS returns cannot be explained by the volatility risk premium only. Market Jump of -5% is the OOPS return from a sudden drop of 5% in the stock market. Jump risk at time t for each option in the portfolio is computed as OtBS (S, σ) − OtBS (0.95S, σ) where OtBS is the Black-Scholes option value for each option in the strategy. We aggregate these exposures in the portfolio and present the result in percent. Market Jump of -5% evolves between −4.8% and 3.7%, and the average time-series value is −0.6%. This exposure is on average negative, i.e, the strategy loses when there are large market drops, but OOPS always loses less than the market. So the magnitude of the exposure is low (and the sign changes often) so that OOPS returns cannot be explained by the jump risk only.27 Therefore, the high OOPS returns cannot be fully explained as compensation for jump risk. From these ex-ante measures it is clear that, on average, OOPS is not particularly exposed to the market, volatility or jump risks. This agrees with the low realized volatility, positive skewness and low kurtosis of the OOPS ex post distribution discussed in the previous section. To clarify the risk exposures of the OOPS even further, we present payoff profiles (the return of the portfolio as a function of S&P500 returns) for the time-series average portfolio 27 We also address more extreme scenarios of market jump. A negative market jump of two (three) standard deviations - a drop of 8% (12%) - results in a average time-series of −2.6% (−6.2%). A positive market jump of of two (three) standard deviations - an increase of 8% (12%) - results in a average time-series of −3.9% (−5.9%). All these are relatively small. We also compute a combination of a large increase in volatility (50% more) and a large drop in the market (a drop of 5%). The average impact on the portfolio is of −2.1% which is relatively small. 20 and three particular dates which present different weight profiles: June 2002, April 2005, and March 2007. Figure 8 presents these four payoff functions in the top four graphs. The OOPS profile in June of 2002 is little exposed to market risk. In April of 2005, OOPS is exposed to low volatility. In March of 2010, OOPS is exposed to left tail risk. These three cases show the diversity of profiles of OOPS. On average, the OOPS time-series average profile is somewhat exposed to the right tail risk. We also present payoff profiles for four traditional strategies for comparison: short OTM put option, short ATM straddle, short zero-beta straddle, and short crash-neutral straddle, all delevered (4.47% weight) so that they can be easily compared to the OOPS. The contrast is stark especially in terms of the volatility exposure. Next, we study the ex post risk exposure of OOPS. Firstly, we define the variables to be used. Excess market return is the difference between market return and the risk-free rate. As a proxy for the market return, we use the S&P 500 index monthly return, rS&P 500 . We proxy jump risk by a variable Jump equal to the value of the S&P 500 index return when the monthly return is below −5% and zero otherwise.28 We consider two volatility measures: realized volatility (RV ) of the previous month using daily data, and the V IX level (V IX). We also construct a proxy for the volatility risk premium, V RP , which is the spread between the implied and realized volatility (e.g. Adrian and Shin (2010), Saretto and Goyal (2009), among others). We use Gamma, the second derivative of the option price in respect to the underlying stock. It proxies for non-linear changes of the stock price. We use the slope of the implied volatility term structure, IV T erm, defined as the 2-month ATM put implied volatility less 1-month ATM put implied volatility. We also use the implied volatility moneyness slope, IV M oneyness, defined as the 1-month OTM put implied volatility less 1-month ATM put implied volatility. All these variables are customary in the equities and 28 We also address more extreme scenarios. We run the same regressions with a market drop of two and three standard deviations. We do the same for a sudden increase of the same magnitude. The conclusions remain unchanged. 21 options literature.29 We run explanatory regressions of out-of-sample OOPS excess returns, rt − rft , on contemporary risk variables, Xt : (15) rt − rft = α + βXt + t We run simple linear regressions on traditional risk factors. The factors considered are excess market return, jump risk, and changes of the following variables: realized volatility, VIX, implied volatility term structure, and implied volatility moneyness slope. Table 7 reports the results. The market coefficient is positive and significant at the 5% level. So OOPS returns are positively related to market returns. The OOPS expected returns are negatively related to contemporaneous jump and changes in realized volatility. All these estimates agree with previous ex ante risk measures findings. Notice that the magnitude of the coefficients on these variables are all low. This shows the low economic risk implied by exposures to market, volatility or jump risks. Moreover, R-squares are very low, i.e, these variables are only able to explain a maximum of 4.99% of the variation of OOPS out of sample returns. Next we study predictive regressions. We regress out–of–sample OOPS excess returns, rt − rft , on a lagged variable Zt−1 , using the following regression: (16) rt − rft = α + βZt−1 + t These regressions study the conditional expected return of the strategy. We therefore use variables that proxy for conditional risk premia. The value at the beginning of the month of the following variables: realized volatility, V IX, variance risk premia, jump risk, gamma, 29 We also tried different explanatory variables and models. For example, there is insignificant exposure to the Carhart factors (http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/), Fung and Hsieh hedge fund factors (http://faculty.fuqua.duke.edu/∼dah7/HFData.htm), liquidity measures such as relative bidask spread or open interest, and the VOL and JUMP factors of Cremers, Halling, and Weinbaum (2015). 22 implied volatility term, and implied volatility moneyness slope. Table 8 shows the results of these regressions. No variable is significant at the 5% significance level. Moreover, these variables present a maximum R2 of 0.41%. We conclude that there is no variable able to meaningfully explain the variation of expected or unexpected OOPS returns. Both the results using ex-ante risk measures and payoff profiles and the ex-post regressions are strong evidence that OOPS returns are mostly based on the exploiting mispricing between options, rather than loading on priced risk factors.30 C Significance Broadie et al. (2009) question the significance of the abnormal returns of simple short option strategies. They show that empirically observed option returns are not inconsistent with what would have been obtained in a Black-Scholes world with no mispricing and no risk premia for volatility or crashes. We conduct a simulation study similar to theirs of a BlackScholes economy under the objective measure. The calibration fits the historic behavior of the S&P500 index returns in our sample between 1950 and 2013. The historic average annualized index return is 7.17%, the annualized volatility is 14.58%, and the annualized risk-free rate is 1.74%. This translates to a historic risk premium of 5.43% and a Sharpe ratio of 0.37 for the stock market, in line with the estimates in Broadie et al. (2009). In this economy there are no transaction costs. We simulate 1, 000 paths of 756 months each (the sample length of our data) of the index level. For each simulated path, we use the first 551 observations as the basis of the simulations used in the OOPS algorithm. These observations corresponds to the initial period between 1950 and 1995 in our sample. For each of the last 212 months, we price options with the moneyness as used in our empirical study using the Black-Scholes formula. Note that, by assumption, in this economy there is no mispricing nor risk premia for crash and volatility risk 30 We only study linear risk factors. We scatter plot the OOPS returns against each risk-factor in order to assess nonlinear dependencies. No pattern is recognized. 23 factors. Then, we run the OOPS algorithm on these artificial samples. In the end we compute the same time-series performance metrics as before, including the certainty equivalent and Sharpe ratio of the OOPS. In Figure 9, we plot the histograms of the difference between simulated OOPS and stock market Sharpe ratios and certainty equivalents, as well as our point estimates reported in Table 4 which are represented by a dashed vertical line. The OOPS annualized Sharpe ratio and certainty equivalent are in the extreme right tail of the simulated distribution showing clearly the significance of our estimates. V Conclusion We offer a new method for portfolio optimization using options. This method takes into account the complex distribution of option returns and investor preferences for higher-order moments. The method is straightforward to implement and requires virtually no computing resources. Our approach does not rely on a long time series of option returns but just requires currently observed prices. The approach takes into account the substantial transaction costs in this market in a simple and convenient manner. We apply the method out-of-sample in the period 1996 to 2013 with impressive results. We obtain a high certainty equivalent and a large Sharpe ratio. We show that these results are driven mostly by exploiting option mispricings, coupled with a small exposure to market and volatility risk premia but not to crash risk. 24 Appendix A In this appendix, we provide a small example of the method explained in Section II.A. In this small example, we assume two time periods, t = 1, 2. At time 1, we run the optimization problem and obtain the weights of each security. At time 2, we perform our out-of-sample exercise. We set the stock prices for this exerise as S1 = 1 for time period 1, and S2 = 0.98 for time period 2 only known at period 2. We also set the risk-free interest rate equal to 1% (rf = 0.01), for both periods. We use four options over the same underlying with the same maturity. The only difference between options is moneyness and the choice between call and put options. Hence, we define the two call options (ATM, c1 , and OTM, c2 ) and two put options (ATM, p1 , and OTM, p2 ) with strike prices given by K1,c1 = K1,p1 = 1, K1,c2 = 1.05 and K1,p2 = 0.95. The current option prices are C1,c1 = 0.04, C1,c2 = 0.0008, P1,p1 = 0.07, P1,p2 = 0.02. We assume a power utility function with γ = 10. In this example, we run our method for 4 simulations and we assume no transaction costs. Table A.1 presents the results for each step of the method explained in Section II.A. We assume the four simulated returns of 5%, 1%, -4%, and -10% in Step 1. Then, in Step 2, we compute the next-period underlying asset value. Using this four values, we compute the simulated optiono payoffs at maturity and the corresponding returns for each option. In Step 4, we compute the simulated portfolio returns. Based on these simulated portfolio returns, we maximize the expected utility function and get the weights for each of the four options. The last step is to determine the out-of-sample portfolio return. The resulting portfolio return for this hypothetical example is of 3.60%. 25 Table A1 Example of the Methodology Table A1 shows an example of the method explained in Section II.A for a two-period model with 4 simulations and assuming no transaction costs. 1. Simulate the underlying asset log-return r21 = 0.0500 r22 = 0.0100 r23 = −0.0400 r24 = −0.1000 2. Find the next-period underlying asset value 1 2 3 = 1.0513 = 1.0101 = 0.9608 S2|1 S2|1 S2|1 4 = 0.9048 S2|1 3.a. Determine simulated option payoffs at maturity 1 2 3 C2|1,c C2|1,c C2|1,c = 0.0513 = 0.0101 = 0.0000 1 1 1 1 2 3 C2|1,c2 = 0.0013 C2|1,c2 = 0.0000 C2|1,c2 = 0.0000 1 2 3 P2|1,p1 = 0.0000 P2|1,p1 = 0.0000 P2|1,p = 0.0392 1 1 2 3 P2|1,p2 = 0.0000 P2|1,p2 = 0.0000 P2|1,p2 = 0.0000 3.b. And corresponding returns for each option 1 2 3 r2|1,c = 0.2818 r2|1,c = −0.7487 r2|1,c = −1.0000 1 1 1 1 2 3 r2|1,c2 = 0.5889 r2|1,c2 = −1.0000 r2|1,c2 = −1.0000 1 2 3 r2|1,p1 = −1.0000 r2|1,p1 = −1.0000 r2|1,p = −0.4398 1 1 2 3 r2|1,p2 = −1.0000 r2|1,p2 = −1.0000 r2|1,p2 = −1.0000 4. Construct the simulated portfolio return rp12|1 = 0.01+ rp22|1 = 0.01+ +ω1,c1 × (0.2718)+ +ω1,c1 × (−0.7587)+ +ω1,c2 × (0.5789)+ +ω1,c2 × (−1.0100)+ +ω1,p1 × (−1.0100)+ +ω1,p1 × (−1.0100)+ +ω1,p2 × (−1.0100) +ω1,p2 × (−1.0100) rp32|1 = 0.01+ +ω1,c1 × (−1.0100)+ +ω1,c2 × (−1.0100)+ +ω1,p1 × (−0.4498)+ +ω1,p2 × (−1.0100) 4 C2|1,c 1 4 C2|1,c 2 4 P2|1,p 1 4 P2|1,p 2 = 0.0000 = 0.0000 = 0.0952 = 0.0452 4 r2|1,c 1 4 r2|1,c 2 4 r2|1,p 1 4 r2|1,p 2 = −1.0000 = −1.0000 = 0.3595 = 1.2581 rp42|1 = 0.01+ +ω1,c1 × (−1.0100)+ +ω1,c2 × (−1.0100)+ +ω1,p1 × (0.3495)+ +ω1,p2 × (1.2481) P (1+rpn )−9 2|1 5. Choose weights by maximizing expected utility over simulated returns 14 4n=1 −9 ω1,c1 = −0.10, ω1,c2 = 0.07, ω1,p1 = −0.05, ω1,p2 = 0.04 and E(U ) = −0.0804 6.a. Determine option actual payoffs... 6.b. ...and returns for each option C2,c1 = 0.0000 C2,c2 = 0.0000 r2,c1 = −1.0000 r2,c2 = −1.0000 P2,p1 = 0.0200 P2,p2 = 0.0000 r2,p1 = −0.7143 r2,p2 = −1.0000 6.c. Determine one-period out-of-sample portfolio return rp2 = 0.0360 26 References Adrian, T., and H. S. Shin.“Liquidity and Leverage.” Journal of Financial Intermediation, 19 (2010), 418–437. Almazan, A.; K. Brown; M. Carlson; and D. Chapman. “Why Constrain your Mutual Fund Manager?” Journal of Financial Economics, 73 (2004), 289–322. Andersen, T. G.; L. Benzoni; and J. Lund. “An Empirical Investigation of Continuous Time Equity Return Models.” Journal of Finance, 57 (2002), 1239–1284. Aragon, G. O., and J. S. Martin. “A Unique View of Hedge Fund Derivatives Usage: Safeguard or Speculation?” Journal of Financial Economics, 105 (2012), 436–456. Baesel, J.; G. Shows; and E. Thorp. “The Cost of Liquidity Services in Listed Options.” Journal of Finance, 38 (1983), 989–995. Bakshi, G.; C. Cao; and Z. Chen. “Empirical Performance of Alternative Option Pricing Models.” Journal of Finance, 52 (1997), 2003–2049. Barone-Adesi, G.; R. F. Engle; and L. Mancini. “A GARCH Option Pricing Model with Filtered Historical Simulation.” Review of Financial Studies, 21 (2008), 1223–1258. Barone-Adesi, G.; K. Giannopolos; and L. Vosper. “VaR without Correlations for NonLinear Portfolios.” Journal of Futures Markets, 19 (1999), 583–602. Bates, D. “Jump and Stochastic Volatility: Exchange Rate Processes Implicit in Deutsche Mark Options.” Review of Financial Studies, 9 (1996), 69–107. Bernardo, A. E., and O. Ledoit. “Gain, Loss, and Asset Pricing.” Journal of Political Economy, 108 (2000), 144–172. Black, F., and M. Scholes. “The Pricing of Options and Corporate Liabilities.” Journal of Political Economy, 81 (1973), 637–659. Bliss, R. R., and N. Panigirtzoglou. “Option-Implied Risk Aversion Estimates.” Journal of Finance, 59 (2004), 407–446. Bondarenko, O. “Why Are Puts so Expensive?” Unpublished working paper, University of 27 Illinois, Chicago (2003). Brandt, M. W. “Estimating Portfolio and Consumption Choice: A Conditional Euler Approach.” Journal of Finance, 54 (1999), 1609–1645. Brandt, M. W.; P. Santa-Clara; and R. Valkanov. “Parametric Portfolio Policies: Exploiting Characteristics in the Cross Section of Equity Returns.” Review of Financial Studies, 22 (2009), 3411–3447. Broadie, M.; M. Chernov; and M. Johannes. “Model Specification and Risk Premia: Evidence from Futures Options.” Journal of Finance, 62 (2007), 1453–1490. Broadie, M.; M. Chernov; and M. Johannes. “Understanding Index Option Returns.” Review of Financial Studies, 22 (2009), 4493–4529. Buraschi, A., and J. Jackwerth. “The Price of a Smile: Hedging and Spanning in Option Markets.” Review of Financial Studies, 14 (2001), 495–527. Chen, Y. “Derivatives Use and Risk Taking: Evidence from the Hedge Fund Industry.” Journal of Financial and Quantitative Analysis, 46 (2011), 1073–1106. Chernov, M., and E. Ghysels. “A Study Towards a Unified Approach to the Joint Estimation of Objective and Risk Neutral Measures for the Purpose of Options Valuation.” Journal of Financial Economics, 56 (2000), 407–458. Christoffersen, P.; K. Jacobs; and S. L. Heston. “Capturing Option Anomalies with a Variance Dependent Pricing Kernel.” Review of Financial Studies 26 (2013), 1963–2006. Constantinides, G. M.; J. C. Jackwerth; and A. Z. Savov. “The Puzzle of Index Option Returns.” The Review of Asset Pricing Studies 3 (2013), 1–29. Coval, J. D., and T. Shumway. “Expected Option Returns.” Journal of Finance 56 (2001), 983–1009. Cremers, M.; M. Halling; and D. Weinbaum. “Aggregate Jump and Volatility Risk in the Cross-Section of Stock Returns.” Journal of Finance Forthcoming (2015). Deli, D. N., and R. Varma. “Contracting in the Investment Management Industry: Evidence from Mutual Funds.” Journal of Financial Economics 63 (2002), 79–98. 28 DeMiguel, V.; L. Garlappi; and R. Uppal. “Optimal versus Naive Diversification: How Inefficient Is the 1/N Portfolio Strategy?” Review of Financial Studies 22 (2009), 1915–1953. Dennis, P., and S. Mayhew. “Microstructural Biases in Empirical Tests of Option Pricing Models.” Review of Derivatives Research 12 (2009), 169–191. Diebold, F. X.; T. Schuermann; and J. D. Stroughair. “Pitfalls and Opportunities in the Use of Extreme Value Theory in Risk Management.” In A.-P. N. Refenes A. N. Burgess & J. D. Moody (Eds.), Decision Technologies for Computational Finance, Kluwer Academic Publishers (1998) 3–12. Driessen, J., and P. Maenhout. “An Empirical Portfolio Perspective on Option Pricing Anomalies.” Review of Finance 11 (2007), 561–603. Driessen, J., and P. Maenhout. “The World Price of Jump and Volatility Risk.” Journal of Banking and Finance 37 (2013), 518–536. Duffie, D.; J. Pan; and K. Singleton. “Transform Analysis and Asset Pricing for Affine Jump-Diffusions.” Econometrica 68 (2000), 1343–1376. Efron, B., and R. J. Tibshirani “An Introduction to the Bootstrap.” Chapman and Hall New York (1993). Eraker, B. “Do Stock Prices and Volatility Jump? Reconciling Evidence from Spot and Option Prices.” Journal of Finance 59 (2004), 1367–1403. Eraker, B. “The Performance of Model Based Option Trading Strategies.” Review of Derivatives Research 16 (2013), 1–23. Hull, J., and A. White. “Incorporating Volatility Updating into the Historical Simulation Method for Value-at-Risk.” Journal of Risk 1 (1998), 5–19. Ingersoll, J.; M. Spiegel; W. Goetzmann; and I. Welch. “Portfolio Performance Manipulation and Manipulation-Proof Performance Measures.” Review of Financial Studies 20 (2007), 1503–1546. Jackwerth, J. “Recovering Risk Aversion from Option Prices and Realized Returns.” Review of Financial Studies 13 (2000), 433–451. 29 Jones, C. “A Nonlinear Factor Analysis of S&P 500 Index Option Returns.” Journal of Finance 61 (2006), 2325–2363. Koski, J. L., and J. Pontiff. “How Are Derivatives Used? Evidence from the Mutual Fund Industry.” Journal of Finance 54 (1999), 791–816. Liu, J., and J. Pan. “Dynamic Derivative Strategies.” Journal of Financial Economics 69 (2003), 401–430. Malamud, S. “Portfolio Selection with Options and Transaction Costs. ”Swiss Finance Institute Research Paper (2014) No. 14-08. Michaud, R., and R. Michaud. “Estimation Error and Portfolio Optimization: A Resampling Solution.” Journal of Investment Management 6 (2008), 8–28. Newey, W. K., and K. D. West. “A simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix.” Econometrica 55 (1987), 703–708. Pan, J. “The Jump-Risk Premia Implicit in Options: Evidence from an Integrated TimeSeries Study.” Journal of Financial Economics 63 (2002), 3–50. Phillips, S., and C. Smith. “Trading Costs for Listed Options: The Implications for Market Efficiency.” Journal of Financial Economics 8 (1980), 179–201. Plyakha, Y., and G. Vilkov. “Portfolio Policies with Stock Options.” Working Paper, (2008). Ross, S. “Options and Efficiency.” Quarterly Journal of Economics 90 (1976), 75–89. Saretto, A., and A. Goyal, 2009, “Cross-Section of Option Returns and Volatility.” Journal of Financial Economics 94 (2009), 310–326. Saretto, A., and P. Santa-Clara. “Option Strategies: Good Deals and Margin Calls.” Journal of Financial Markets 12 (2009), 391–417. Tan, S. “The Role of Options in Long Horizon Portfolio Choice.” The Journal of Derivatives 20 (2013), 60–77. Todorov, V. “Variance Risk Premium Dynamics: The Role of Jumps.” Review of Financial Studies 23 (2010), 345–383. Vanden, J. “Options Trading and the CAPM.” Review of Financial Studies 17 (2004), 30 207–238. 31 Figure 1 Bid-Ask Spread Figure 1 represents monthly observations of the relative bid-ask spread (the ratio of the absolute bid-ask spread to the mid price) for the period between January 1996 and August 2013 for four options: a ATM call, a ATM put, a 5% OTM call, and a 5% OTM put options. Each option security was created as defined in Section 1. Results in percentage. Period: January 1996 to August 2013. 40 ATM call ATM put OTM call OTM put 35 30 % 25 20 15 10 5 0 1996 1998 2000 2002 2004 2006 Year 32 2008 2010 2012 Figure 2 S&P 500 Index and VIX Figure 2 represents monthly observations of the S&P 500 total return index and the VIX index. Period: January 1996 to August 2013. 2000 100 1000 0 1996 50 1998 2000 2002 2004 2006 Year 33 2008 2010 2012 0 VIX (%) S&P 500 Index S&P 500 Index VIX Figure 3 Densities of Monthly Option Returns Figure 3 represents the histograms of monthly raw returns of a long position in ATM and 5% OTM options over the S&P 500 index. Period: January 1996 to August 2013. Graph B. ATM Put -100 Frequency Frequency Graph A. ATM Call 0 100 200 300 400 500 Returns (%) 600 700 800 -100 900 0 300 400 500 Returns (%) 600 700 800 900 800 900 Graph D. OTM Put Frequency -100 200 Frequency Graph C. OTM Call 100 0 100 200 300 400 500 Returns (%) 600 700 800 900 -100 34 0 100 200 300 400 500 Returns (%) 600 700 Figure 4 OOPS Return Distribution Figure 4 presents the distribution of out-of-sample one-month OOPS returns. OOPS returns Frequency are after transaction costs. Period: January 1996 to August 2013. -25 -20 -15 -10 -5 0 Returns (%) 35 5 10 15 20 25 Figure 5 OOPS Cumulative Returns Figure 5 presents out-of-sample one-month OOPS cumulative returns. OOPS returns are after transaction costs. Period: January 1996 to August 2013. 2000 OOPS S&P 500 Risk-free 1800 1600 1400 1200 1000 800 600 400 200 0 1996 1998 2000 2002 2004 2006 Year 36 2008 2010 2012 Figure 6 OOPS Time-Series Weights Figure 6 shows the monthly weights of OOPS. The top left panel presents weights for the ATM call and put options. The top right panel presents weights for the 5% OTM call and put options. The bottom left panel presents weights for the difference between ATM options and OTM options. The bottom right panel presents the weight of the risk-free security. Period: January 1996 to August 2013. Graph A. ATM Call and Put Graph B. OTM Call and Put 10 10 OTM call ATM put 8 6 6 4 4 2 2 Weight (%) Weight (%) ATM call 8 0 -2 0 -2 -4 -4 -6 -6 -8 -8 -10 1996 1998 2000 2002 2004 2006 Year 2008 2010 -10 1996 2012 OTM put Graph C. ATM C-P and OTM C-P 1998 2000 2002 2004 2006 Year 2008 2010 2012 Graph D. Risk-Free 120 10 ATM C-P OTM C-P 8 115 6 110 2 Weight (%) Weight (%) 4 0 -2 105 100 -4 -6 95 -8 -10 1996 1998 2000 2002 2004 2006 Year 2008 2010 90 1996 2012 37 1998 2000 2002 2004 2006 Year 2008 2010 2012 Figure 7 OOPS Time-Series Risk Measures Figure 7 presents OOPS risk measures calculated according to the Black-Scholes model. Beta is the percentage change in the portfolio to a one percent change in the underlying price. Percent Vega is the percentage change in the portfolio due to a one percentage point change in the volatility. Market Jump of -5% is the percentage change in portfolio if the stock market drops 5%. Period: January 1996 to August 2013. Graph A. Beta 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1 1996 1998 2000 2002 2004 2006 Year 2008 2010 2012 Graph B. Percent Vega 2 1.5 1 % 0.5 0 -0.5 -1 -1.5 -2 1996 1998 2000 2002 2004 2006 Year 2008 2010 2012 Graph C. Market Jump of -5% 5 4 3 2 % 1 0 -1 -2 -3 -4 -5 1996 1998 2000 2002 2004 2006 Year 38 2008 2010 2012 Figure 8 Payoff Profile Figure 8 presents the effect of a percentage change of the stock market on several strategies. The first four top graphs are the time-series average and three specific dates for the OOPS allocation and the four bottom graphs represents traditional short strategies used in the literature related to market, volatility and jump risk. Period: January 1996 to August 2013. Graph B. OOPS – June 2002 20 15 15 10 10 5 5 Return (%) Return (%) Graph A. OOPS – Average 20 0 -5 -10 -10 -15 -15 -20 -10 -5 0 Change in S&P 500 (%) 5 -20 -10 10 20 15 15 10 10 5 5 0 -5 -10 -10 -5 0 Change in S&P 500 (%) 5 -20 -10 10 10 -5 0 Change in S&P 500 (%) 5 10 Graph F. Short ATM Straddle 20 15 15 10 10 5 5 Return (%) Return (%) Graph E. Short OTM Put 20 0 0 -5 -5 -10 -10 -15 -15 -5 0 Change in S&P 500 (%) 5 -20 -10 10 Graph G. Short Zero-Beta Straddle -5 0 Change in S&P 500 (%) 5 10 Graph H. Short Crash Straddle 20 20 15 10 10 5 5 Return (%) 15 0 0 -5 -5 -10 -10 -15 -20 -10 5 -15 -15 -20 -10 0 Change in S&P 500 (%) 0 -5 -20 -10 -5 Graph D. OOPS – March 2007 20 Return (%) Return (%) Graph C. OOPS – April 2005 Return (%) 0 -5 -15 -5 0 Change in S&P 500 (%) 5 -20 -10 10 39 -5 0 Change in S&P 500 (%) 5 10 Figure 9 Sampling Distribution of OOPS Sharpe Ratio and Certainty Equivalent Under Black-Scholes Economy Figure 9 presents histograms of the difference between the simulated OOPS and simulated stock market Sharpe ratios and the difference of the simulated OOPS and simulated stock market certainty equivalents under a Black-Scholes economy. The experiment uses 1,000 trajectories. The solid vertical line is the observed value from the data. Graph A. SR(OOPS) – SR(Stock Market) SR(OOPS)-SR(Stock Market) -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 Graph B. CE(OOPS) – CE(Stock Market) CE(OOPS)-CE(Stock Market) -25 -20 -15 -10 -5 % 40 0 5 10 15 Table 1 S&P 500 Index Returns – Summary Statistics Table 1 reports summary statistics (number of observations, skewness, excess kurtosis, one-month autocorrelation for the returns and squared returns) and tests (one-month Ljung–Box and Arch tests; p-values of each test are presented in squared brackets) for raw and standardized returns of S&P 500 index. Obs Skew Exc Kurt ρ1 (z) ρ1 (z 2 ) Q1 (z) Arch(1) 1950-1995 551 −0.56 5.97 0.03 0.13 0.44 [0.51] 9.60 [0.00] Raw returns 1996-2013 1950-2013 212 763 −0.81 −0.66 4.30 5.41 0.10 0.05 0.23 0.16 2.07 2.08 [0.15] [0.15] 10.83 20.50 [0.00] [0.00] 41 Standardized returns 1950-1995 1996-2013 1950-2013 551 212 763 −0.34 −0.48 −0.35 3.60 3.36 3.78 0.03 0.04 0.03 −0.08 −0.10 −0.06 0.38 0.27 0.64 [0.54] [0.60] [0.42] 3.93 2.04 3.12 [0.05] [0.15] [0.08] Table 2 Summary Statistics of Options Table 2 reports averages of option moneyness, price, bid-ask spread, relative bid-ask spread, volume, open interest (in thousands), implied volatility, delta, beta, gamma and vega for four options with one-month to maturity: a ATM call, a ATM put, a 5% OTM call, and a 5% OTM put options using mid prices. Option moneyness is defined as S/K -1. Price is the mid option price. Bid-ask spread is the difference between ask and bid price. Relative bid-ask spread is the ratio of the bid-ask spread to the mid price. Volume is the contract’s volume on the day when there is one-month to expiration. Open interest is the open interest prevalent on the same day. Implied volatility is the annualized volatility of the option using the Black-Scholes model. Delta is the Black-Scholes delta. Beta is the product of delta and the ratio of underlying asset value and option price. Gamma is the Black-Scholes gamma. Vega is the Black-Scholes vega. Period: January 1996 to August 2013. Option moneyness Price Bid-ask spread Relative bid-ask spread Volume Open interest ($000) Implied volatility Delta Beta Gamma Vega ATM Call 0.24% 27.31 1.47 5.22% 12,298 15,356 19.48% 0.53 25.32 0.01 1.33 42 ATM Put OTM Call OTM Put −0.10% −4.09% 4.37% 24.97 6.79 10.86 1.37 0.65 0.84 5.27% 10.91% 7.68% 13,322 13,092 17,342 15,756 15,721 15,591 18.41% 16.89% 21.83% −0.50 0.19 −0.23 −26.57 47.33 −30.60 0.01 0.01 0.01 1.34 0.84 0.99 Table 3 Summary Statistics of Returns Table 3 reports summary statistics (annualized mean, annualized standard deviation, minimum, maximum, skewness, excess kurtosis, annualized certainty equivalent, and annualized Sharpe ratio) for a buy-and-hold strategy in several assets: primitives assets (S&P 500 total return index and one-month T-Bill), options (a ATM call, a ATM put, a 5% OTM call and a 5% OTM put), a strategy that allocates a weight of 1/4 to each one of the options previously selected), and delevered options composed by one of the previous options with weight of −4.47% and 104.47% in the risk-free asset. Period: January 1996 to August 2013. Ann Ann Mean Std Dev Min Max 6.4% 1.1% ATM call ATM put OTM call OTM put 1/N Rule 52.4% −209.7% −131.3% −440.2% −182.2% 415.6% 528.9% 947.0% 738.0% 390.0% ATM call ATM put OTM call OTM put 1/N Rule −1.2% 10.5% 7.0% 20.8% 9.2% 18.6% 23.6% 42.3% 33.0% 17.4% Exc Kurt Ann CE Ann SR Primitive assets 18.6% −1.24 5.57 −1.31% 0.29 0.2% 0.07 −1.59 Only options −100% 409% 0.91 −0.12 −100.00% 0.12 −100% 703% 2.44 6.28 −100.00% −0.40 −100% 2,496% 5.38 38.17 −100.00% −0.14 −100% 1,265% 4.15 17.88 −100.00% −0.60 −100% 664% 2.73 9.14 −100.00% −0.47 Delevered options −18% 5% −0.91 −0.13 −8.42% −0.12 −31% 5% −2.44 6.30 −4.93% 0.40 −111% 5% −5.38 38.16 −100.00% 0.14 −57% 5% −4.16 17.90 −31.78% 0.60 −30% 5% −2.72 9.12 1.32% 0.47 18.2% −29.5% 0.2% 0.0% S&P 500 1m T-Bill Skew 43 Table 4 Out-of-Sample OOPS Returns Table 4 reports time-series summary statistics (annualized mean, annualized standard deviation, minimum, maximum, skewness, excess kurtosis, annualized certainty equivalent, and annualized Sharpe ratio) for the S&P 500 total return index and the out-of-sample OOPS returns as explained in Section 1. OOPS returns are after transaction costs. Period: January 1996 to August 2013. Ann Ann Min Mean Std Dev S&P 500 6.4% 18.2% −29.5% OOPS 16.1% 18.4% −18.4% Max Skew 18.6% −1.24 20.8% 0.35 44 Exc Kurt 5.57 2.03 Ann Ann CE SR −1.31% 0.29 9.94% 0.82 Table 5 Out-of-Sample OOPS Returns – Different Volatility Estimators Table 5 reports time-series summary statistics (annualized mean, annualized standard deviation, minimum, maximum, skewness, excess kurtosis, annualized certainty equivalent, and annualized Sharpe ratio) for the out-of-sample OOPS returns for different estimators of volatility. We present in brackets the estimator used. RV stands for realized volatility, VIX stands for VIX, and RM stands for RiskMetrics. OOPS returns are after transaction costs. Period: January 1996 to August 2013. OOPS (RV) Optimal d d=1 d=5 d = 10 d = 20 d = 30 OOPS (RM) λ = 0.80 λ = 0.90 λ = 0.96 OOPS (VIX) Ann Mean Ann Std Dev Min Max Skew Exc Kurt Ann CE Ann SR 16.1% 11.1% 11.9% 7.8% 3.7% 2.1% 18.4% 14.4% 13.8% 12.1% 11.1% 9.5% −18.4% −11.6% −12.9% −15.2% −18.4% −15.2% 20.8% 20.8% 20.4% 16.1% 11.7% 7.6% 0.35 2.04 0.93 0.07 −1.52 −2.06 2.03 7.44 4.47 4.89 8.59 9.06 9.94% 8.23% 8.65% 4.97% 0.95% 0.05% 0.82 0.75 0.79 0.56 0.23 0.11 9.3% 5.9% 3.7% 6.6% 12.1% 10.1% 8.8% 11.2% −13.9% −15.2% −14.5% −16.8% 19.6% 0.61 14.9% − 0.41 9.9% −1.38 16.0% −0.33 6.11 8.33 9.57 7.75 6.61% 3.87% 2.12% 4.03% 0.68 0.48 0.31 0.49 45 Table 6 OOPS Weights Table 6 presents time-series mean, minimum and maximum weights for each risky security of the OOPS. Period: January 1996 to August 2013. ATM Call Mean −2.5% Min −8.7% Max 1.8% ATM Put −1.8% −8.4% 3.2% 46 OTM Call 0.2% −3.8% 3.5% OTM Put −0.4% −4.2% 0.9% Table 7 Explanatory Regressions for OOPS Returns Table 7 presents estimated coefficients (p-values in squared brackets) and the adjusted R2 for explanatory regressions in which the dependent variable is the out-of-sample OOPS excess returns and the independent variable is the variable stated in the left column and defined in Section 3. ***, **, * denotes significant estimated parameters at 1%, 5%, and 10% levels, respectively. Period: January 1996 to August 2013 corresponding to 212 mothly observations. Variable rS&P 500 − rf OOPS excess returns 0.14∗∗ [0.04] 0.34∗∗∗ [0.00] Jump −0.07 [0.67] ∆RV −0.14 [0.51] ∆V IX −0.14 [0.49] ∆IV T erm ∆IV M oneyness 2 R (%) 1.46 4.99 47 −0.06 [0.87] −0.39 −0.27 −0.24 −0.47 Table 8 Predictive Regressions for OOPS Returns Table 8 presents estimated coefficients (p-values in squared brackets) and the adjusted R2 for predictive regressions in which the dependent variable is the out-of-sample OOPS excess returns and the independent variable is the variable stated in the left column and defined in Section 3. ***, **, * denotes significant estimated parameters at 1%, 5%, and 10% levels, respectively. Period: January 1996 to August 2013 corresponding to 212 mothly observations. Variable RV OOPS excess returns 0.07 [0.60] V IX 0.17 [0.22] 0.32 [0.22] V RP −0.03 [0.78] Jump −0.13 [0.17] Gamma IV T erm −0.08 [0.73] IV M oneyness −0.07 [0.85] −0.42 −0.46 2 R (%) −0.35 0.24 0.23 48 −0.44 0.41